D4: the PARP1 natural-product hit list came back empty — as pre-registered

What we expected

Re-scoring 60 previously-flagged compounds under the newly-fixed pocket constraint (threshold moved from 0.500 to 0.897) would confirm or reduce a prior natural-product hit list for PARP1.

What happened

All 60 compounds cleared the old 0.500 threshold; only 17 cleared the new constrained threshold of 0.897. Of those 17, fifteen were reference PARP inhibitors, all with strongly negative affinity (−0.99 to −2.61):

• pamiparib 0.994
• fluzoparib 0.992
• rucaparib 0.986
• niraparib 0.978
• veliparib 0.977
• stenoparib 0.968
• talazoparib 0.967
• PJ-34 0.957

The remaining two were natural products, both scoring 0.898 — clearing the 0.897 bar by 0.001.

Why it happened

The two natural-product “hits” failed on three independent grounds: (1) the 0.001 margin is inside the 0.004–0.010 replicate noise of the bar itself; (2) the affinity head disagreed with the binary head — both natural products had positive affinity (+0.59, +0.50) where every reference drug was negative, which is what a false positive of the binary head looks like; and (3) both were the pyrrolizidine alkaloid N-oxide scaffold, which the natural-products review had already flagged for exclusion as genotoxic and hepatocarcinogenic.

What changed

The pre-registered plan (written 2026-08-14) stated that zero is a legitimate, reportable outcome, and that the gates must not be relaxed to produce a number — and zero is what was reported. The real driver of the change was the threshold moving from 0.500 (uncalibrated) to 0.897 (from the properly constrained recalibration): at 0.500, all 60 compounds looked like hits.

“D4 — How many NP hits survive all five controls? Zero is a legitimate, reportable outcome. The 2026-06-14 audit precedent was 0 of 17. Do not relax the gates to produce a number.”